4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * This file contains the default values for the operation of the
9 * Linux VM subsystem. Fine-tuning documentation can be found in
10 * Documentation/sysctl/vm.txt.
12 * Swap aging added 23.2.95, Stephen Tweedie.
13 * Buffermem limits added 12.3.98, Rik van Riel.
17 #include <linux/sched.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/swap.h>
20 #include <linux/mman.h>
21 #include <linux/pagemap.h>
22 #include <linux/pagevec.h>
23 #include <linux/init.h>
24 #include <linux/module.h>
25 #include <linux/mm_inline.h>
26 #include <linux/buffer_head.h> /* for try_to_release_page() */
27 #include <linux/percpu_counter.h>
28 #include <linux/percpu.h>
29 #include <linux/cpu.h>
30 #include <linux/notifier.h>
31 #include <linux/backing-dev.h>
32 #include <linux/memcontrol.h>
34 /* How many pages do we try to swap or page in/out together? */
37 static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
38 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
41 * This path almost never happens for VM activity - pages are normally
42 * freed via pagevecs. But it gets used by networking.
44 static void __page_cache_release(struct page *page)
48 struct zone *zone = page_zone(page);
50 spin_lock_irqsave(&zone->lru_lock, flags);
51 VM_BUG_ON(!PageLRU(page));
53 del_page_from_lru(zone, page);
54 spin_unlock_irqrestore(&zone->lru_lock, flags);
59 static void put_compound_page(struct page *page)
61 page = compound_head(page);
62 if (put_page_testzero(page)) {
63 compound_page_dtor *dtor;
65 dtor = get_compound_page_dtor(page);
70 void put_page(struct page *page)
72 if (unlikely(PageCompound(page)))
73 put_compound_page(page);
74 else if (put_page_testzero(page))
75 __page_cache_release(page);
77 EXPORT_SYMBOL(put_page);
80 * put_pages_list() - release a list of pages
81 * @pages: list of pages threaded on page->lru
83 * Release a list of pages which are strung together on page.lru. Currently
84 * used by read_cache_pages() and related error recovery code.
86 void put_pages_list(struct list_head *pages)
88 while (!list_empty(pages)) {
91 victim = list_entry(pages->prev, struct page, lru);
92 list_del(&victim->lru);
93 page_cache_release(victim);
96 EXPORT_SYMBOL(put_pages_list);
99 * pagevec_move_tail() must be called with IRQ disabled.
100 * Otherwise this may cause nasty races.
102 static void pagevec_move_tail(struct pagevec *pvec)
106 struct zone *zone = NULL;
108 for (i = 0; i < pagevec_count(pvec); i++) {
109 struct page *page = pvec->pages[i];
110 struct zone *pagezone = page_zone(page);
112 if (pagezone != zone) {
114 spin_unlock(&zone->lru_lock);
116 spin_lock(&zone->lru_lock);
118 if (PageLRU(page) && !PageActive(page)) {
119 int lru = page_is_file_cache(page);
120 list_move_tail(&page->lru, &zone->lru[lru].list);
125 spin_unlock(&zone->lru_lock);
126 __count_vm_events(PGROTATED, pgmoved);
127 release_pages(pvec->pages, pvec->nr, pvec->cold);
128 pagevec_reinit(pvec);
132 * Writeback is about to end against a page which has been marked for immediate
133 * reclaim. If it still appears to be reclaimable, move it to the tail of the
136 void rotate_reclaimable_page(struct page *page)
138 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
140 struct pagevec *pvec;
143 page_cache_get(page);
144 local_irq_save(flags);
145 pvec = &__get_cpu_var(lru_rotate_pvecs);
146 if (!pagevec_add(pvec, page))
147 pagevec_move_tail(pvec);
148 local_irq_restore(flags);
153 * FIXME: speed this up?
155 void activate_page(struct page *page)
157 struct zone *zone = page_zone(page);
159 spin_lock_irq(&zone->lru_lock);
160 if (PageLRU(page) && !PageActive(page)) {
161 int file = page_is_file_cache(page);
162 int lru = LRU_BASE + file;
163 del_page_from_lru_list(zone, page, lru);
167 add_page_to_lru_list(zone, page, lru);
168 __count_vm_event(PGACTIVATE);
169 mem_cgroup_move_lists(page, true);
171 zone->recent_rotated[!!file]++;
172 zone->recent_scanned[!!file]++;
174 spin_unlock_irq(&zone->lru_lock);
178 * Mark a page as having seen activity.
180 * inactive,unreferenced -> inactive,referenced
181 * inactive,referenced -> active,unreferenced
182 * active,unreferenced -> active,referenced
184 void mark_page_accessed(struct page *page)
186 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
188 ClearPageReferenced(page);
189 } else if (!PageReferenced(page)) {
190 SetPageReferenced(page);
194 EXPORT_SYMBOL(mark_page_accessed);
196 void __lru_cache_add(struct page *page, enum lru_list lru)
198 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
200 page_cache_get(page);
201 if (!pagevec_add(pvec, page))
202 ____pagevec_lru_add(pvec, lru);
203 put_cpu_var(lru_add_pvecs);
207 * lru_cache_add_lru - add a page to a page list
208 * @page: the page to be added to the LRU.
209 * @lru: the LRU list to which the page is added.
211 void lru_cache_add_lru(struct page *page, enum lru_list lru)
213 if (PageActive(page)) {
214 ClearPageActive(page);
217 VM_BUG_ON(PageLRU(page) || PageActive(page));
218 __lru_cache_add(page, lru);
222 * Drain pages out of the cpu's pagevecs.
223 * Either "cpu" is the current CPU, and preemption has already been
224 * disabled; or "cpu" is being hot-unplugged, and is already dead.
226 static void drain_cpu_pagevecs(int cpu)
228 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
229 struct pagevec *pvec;
233 pvec = &pvecs[lru - LRU_BASE];
234 if (pagevec_count(pvec))
235 ____pagevec_lru_add(pvec, lru);
238 pvec = &per_cpu(lru_rotate_pvecs, cpu);
239 if (pagevec_count(pvec)) {
242 /* No harm done if a racing interrupt already did this */
243 local_irq_save(flags);
244 pagevec_move_tail(pvec);
245 local_irq_restore(flags);
249 void lru_add_drain(void)
251 drain_cpu_pagevecs(get_cpu());
256 static void lru_add_drain_per_cpu(struct work_struct *dummy)
262 * Returns 0 for success
264 int lru_add_drain_all(void)
266 return schedule_on_each_cpu(lru_add_drain_per_cpu);
272 * Returns 0 for success
274 int lru_add_drain_all(void)
282 * Batched page_cache_release(). Decrement the reference count on all the
283 * passed pages. If it fell to zero then remove the page from the LRU and
286 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
287 * for the remainder of the operation.
289 * The locking in this function is against shrink_inactive_list(): we recheck
290 * the page count inside the lock to see whether shrink_inactive_list()
291 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
294 void release_pages(struct page **pages, int nr, int cold)
297 struct pagevec pages_to_free;
298 struct zone *zone = NULL;
299 unsigned long uninitialized_var(flags);
301 pagevec_init(&pages_to_free, cold);
302 for (i = 0; i < nr; i++) {
303 struct page *page = pages[i];
305 if (unlikely(PageCompound(page))) {
307 spin_unlock_irqrestore(&zone->lru_lock, flags);
310 put_compound_page(page);
314 if (!put_page_testzero(page))
318 struct zone *pagezone = page_zone(page);
319 if (pagezone != zone) {
321 spin_unlock_irqrestore(&zone->lru_lock,
324 spin_lock_irqsave(&zone->lru_lock, flags);
326 VM_BUG_ON(!PageLRU(page));
327 __ClearPageLRU(page);
328 del_page_from_lru(zone, page);
331 if (!pagevec_add(&pages_to_free, page)) {
333 spin_unlock_irqrestore(&zone->lru_lock, flags);
336 __pagevec_free(&pages_to_free);
337 pagevec_reinit(&pages_to_free);
341 spin_unlock_irqrestore(&zone->lru_lock, flags);
343 pagevec_free(&pages_to_free);
347 * The pages which we're about to release may be in the deferred lru-addition
348 * queues. That would prevent them from really being freed right now. That's
349 * OK from a correctness point of view but is inefficient - those pages may be
350 * cache-warm and we want to give them back to the page allocator ASAP.
352 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
353 * and __pagevec_lru_add_active() call release_pages() directly to avoid
356 void __pagevec_release(struct pagevec *pvec)
359 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
360 pagevec_reinit(pvec);
363 EXPORT_SYMBOL(__pagevec_release);
366 * pagevec_release() for pages which are known to not be on the LRU
368 * This function reinitialises the caller's pagevec.
370 void __pagevec_release_nonlru(struct pagevec *pvec)
373 struct pagevec pages_to_free;
375 pagevec_init(&pages_to_free, pvec->cold);
376 for (i = 0; i < pagevec_count(pvec); i++) {
377 struct page *page = pvec->pages[i];
379 VM_BUG_ON(PageLRU(page));
380 if (put_page_testzero(page))
381 pagevec_add(&pages_to_free, page);
383 pagevec_free(&pages_to_free);
384 pagevec_reinit(pvec);
388 * Add the passed pages to the LRU, then drop the caller's refcount
389 * on them. Reinitialises the caller's pagevec.
391 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
394 struct zone *zone = NULL;
396 for (i = 0; i < pagevec_count(pvec); i++) {
397 struct page *page = pvec->pages[i];
398 struct zone *pagezone = page_zone(page);
400 if (pagezone != zone) {
402 spin_unlock_irq(&zone->lru_lock);
404 spin_lock_irq(&zone->lru_lock);
406 VM_BUG_ON(PageLRU(page));
408 if (is_active_lru(lru))
410 add_page_to_lru_list(zone, page, lru);
413 spin_unlock_irq(&zone->lru_lock);
414 release_pages(pvec->pages, pvec->nr, pvec->cold);
415 pagevec_reinit(pvec);
418 EXPORT_SYMBOL(____pagevec_lru_add);
421 * Try to drop buffers from the pages in a pagevec
423 void pagevec_strip(struct pagevec *pvec)
427 for (i = 0; i < pagevec_count(pvec); i++) {
428 struct page *page = pvec->pages[i];
430 if (PagePrivate(page) && trylock_page(page)) {
431 if (PagePrivate(page))
432 try_to_release_page(page, 0);
439 * pagevec_swap_free - try to free swap space from the pages in a pagevec
440 * @pvec: pagevec with swapcache pages to free the swap space of
442 * The caller needs to hold an extra reference to each page and
443 * not hold the page lock on the pages. This function uses a
444 * trylock on the page lock so it may not always free the swap
445 * space associated with a page.
447 void pagevec_swap_free(struct pagevec *pvec)
451 for (i = 0; i < pagevec_count(pvec); i++) {
452 struct page *page = pvec->pages[i];
454 if (PageSwapCache(page) && trylock_page(page)) {
455 if (PageSwapCache(page))
456 remove_exclusive_swap_page_ref(page);
463 * pagevec_lookup - gang pagecache lookup
464 * @pvec: Where the resulting pages are placed
465 * @mapping: The address_space to search
466 * @start: The starting page index
467 * @nr_pages: The maximum number of pages
469 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
470 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
471 * reference against the pages in @pvec.
473 * The search returns a group of mapping-contiguous pages with ascending
474 * indexes. There may be holes in the indices due to not-present pages.
476 * pagevec_lookup() returns the number of pages which were found.
478 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
479 pgoff_t start, unsigned nr_pages)
481 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
482 return pagevec_count(pvec);
485 EXPORT_SYMBOL(pagevec_lookup);
487 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
488 pgoff_t *index, int tag, unsigned nr_pages)
490 pvec->nr = find_get_pages_tag(mapping, index, tag,
491 nr_pages, pvec->pages);
492 return pagevec_count(pvec);
495 EXPORT_SYMBOL(pagevec_lookup_tag);
499 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
502 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
504 static DEFINE_PER_CPU(long, committed_space);
506 void vm_acct_memory(long pages)
511 local = &__get_cpu_var(committed_space);
513 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
514 atomic_long_add(*local, &vm_committed_space);
520 #ifdef CONFIG_HOTPLUG_CPU
522 /* Drop the CPU's cached committed space back into the central pool. */
523 static int cpu_swap_callback(struct notifier_block *nfb,
524 unsigned long action,
529 committed = &per_cpu(committed_space, (long)hcpu);
530 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
531 atomic_long_add(*committed, &vm_committed_space);
533 drain_cpu_pagevecs((long)hcpu);
537 #endif /* CONFIG_HOTPLUG_CPU */
538 #endif /* CONFIG_SMP */
541 * Perform any setup for the swap system
543 void __init swap_setup(void)
545 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
548 bdi_init(swapper_space.backing_dev_info);
551 /* Use a smaller cluster for small-memory machines */
557 * Right now other parts of the system means that we
558 * _really_ don't want to cluster much more
560 #ifdef CONFIG_HOTPLUG_CPU
561 hotcpu_notifier(cpu_swap_callback, 0);